Student Dropout Prediction

Abstract

In India, student dropout is a major issue, especially in the case of students attending private schools, which are not affordable for the majority of the population. This affects not only the distribution of education but also the whole country’s development. Therefore, the paper suggests the use of a machine learning-based system for early warning that would help in predicting which students might drop out by using several indicators such as academic performance, attendance, behavior, socio-economic status, and digital engagement. A diverse dataset was created by combining institutional student records with open- access educational datasets, and the data was preprocessed via normalization, encoding, and feature engineering techniques. A number of predictive models such as Logistic Regression, Support Vector Machine (SVM), Random Forest, Gradient Boosting, and XGBoost were trained and their performance evaluated with respect to accuracy, precision, recall, and F1-score. The results indicated XGBoost as the best model, achieving an accuracy of 93%, a precision of 91%, a recall of 92%, and an F1- score of 91%, which is higher than that of the other baseline models. The system proposed makes it possible to detect these students at risk of failing in a proactive manner, which in turn facilitates timely academic, financial, and emotional support through the existing institution’s mechanisms. The results prove that the use of ensemble learning techniques is very productive when it comes to dealing with complex educational data and can greatly improve the effectiveness of data-driven student retention strategies in educational institutions